25 Oct
2011
25 Oct
'11
12:47 p.m.
On Mon, Oct 24, 2011 at 10:33:13PM -0400, Paul Davis wrote:
i'm relatively skeptical of any claims for audio
steganography that
can survive arbitrary psycho-acoustic compression.
It can still be done, using 'high level' features of the audio
signal. For example, a lot of music contains regular rythmic
patterns. Small timing errors can be used to insert a watermark.
Even if the result sounds different when compared to the original,
that doesn't mean you could tell which is which - the watermarked
file can be 'as good' as the original.
And in fact in many cases it's not even necessary to do this: the
content watermarks itself. The probability that another rendering
of the same music would reproduce the same error pattern is as
good as zero. So if you find this pattern the file must be a copy.
A variation on this is used to check the authenticity of some
recordings in forensic audio. Almost all 'surveillance' audio
contains some low level 50/60 Hz. If the recording is done on
a digital device then its clock can be assumed to be more stable
than the mains frequency, so any variations in this can be
measured and compared against a database (which some forensic
labs have been building up over the last years - it's possible
because power grids are coupled over very large areas, e.g.
almost all of western Europe). The result indicates when the
recording was made. 10-15 minutes is enough for a very reliable
timestamp in most cases.
in addition, random
permutations of the least significant bits of a PCM encoding will
almost certainly eliminate or at least reduce the confidence level
associated with the presence or absence of the watermark), in a way
that will be inaudible to more or less anyone.
Watermarks are not inserted directly as 'digital' information,
(as could be done for steganography) precisely for this reason.
Ciao,
--
FA